I’ve been using the bot to generate texts (a mixture of fiction and non-fiction) at an “Intermediate” level in both Māori and Russian. I’m noticing that there is some pretty wide variability in the CEFR level they get in the Library, though.
For Māori, they range all the way from A2 to C1 (even though the pre-set is consistently ‘intermediate’.)
For Russian, I’ve only tried a small handful, but they’re all consistently graded “C2”.
It’s difficult to say with certainty with two such different languages, but I wouldn’t say either set of texts is ‘easier’ (or harder) for me to read, and I’d consider my knowledge of both to be ‘intermediate’, so in that sense, I see them as being similar in difficulty, but the bot clearly does not
I noticed a similar thing for Finnish texts. Almost all generated texts are graded “C2” even if I set the level to “Beginner”. And it seems that most Finnish text in the library are graded “C2”, but some of them are definitely much easier than others.
Fwiw, I’ve just taken to generating them in ChatGPT directly and then uploading them here. It’s (so far) pretty passable at distinguishing between, say, B2 and C1 (rather than just beginner/intermediate/advanced)
I must warn you that the CEFR levels (A1…C2) that Readlang reports should be taken with a pinch of salt. At some point I should probably revisit how these are calculated and perhaps indicate more clearly in the UI that they are only a rough estimate.
